Projects

### Overview: Lc0 is a cutting-edge neural network chess engine designed by the LeelaChessZero project. It leverages the power of GPUs for enhanced performance and supports a wide range of hardware configurations, making it a versatile tool for both training and playing chess. ### Key Features: - **Neural Network Engine**: Utilizes advanced neural networks to analyze chess positions and make moves. - **GPU Acceleration**: Employs GPUs for significant speed improvements, making it highly efficient for both training and playing. - **Cross-Platform Support**: Compatible with various hardware configurations, including CPUs and GPUs, ensuring broad accessibility. ### Tech Stack: - **C++**: The primary programming language for the engine. - **Neural Networks**: Powered by cutting-edge neural network architectures. - **CUDA/ROCm/Intel DNNL**: GPU acceleration libraries for enhanced performance. - **Ninja**: Build system for efficient project management. ### Why It Matters: Lc0 is crucial for researchers and enthusiasts alike, offering a powerful tool for improving chess AI and exploring the boundaries of neural network applications in the field of computer science. Its ability to leverage GPU acceleration makes it highly competitive in both training and playing scenarios, setting it apart from traditional AI approaches.

Overview: Harbor is a comprehensive AI stack solution that simplifies the configuration of your AI infrastructure with a single command. Key features: - Pre-wired LLM stack - Hundreds of services included - One command to bring everything together Tech stack: Python Why it matters: Harbor simplifies the setup of AI infrastructure, making it easier to deploy and manage without the need for extensive configuration. Repository: av/harbor Primary language: Python GitHub stars: 3248 GitHub description: Stop configuring your AI stack. Start using it. One command brings a complete pre-wired LLM stack with hundreds of services to explore.

### Overview: OneTrainer is a comprehensive solution for Diffusion training, offering a wide range of features and support for various models and formats. ### Key Features: - **Supported Models**: Ernie Image, Z-Image, Qwen Image, FLUX.1, Flux.2 Dev and Klein, Chroma, Stable Diffusion 1.5, 2.0, 2.1, 3.0, 3.5, SDXL, Würstchen-v2, Stable Cascade, PixArt-Alpha, PixArt-Sigma, Sana, Hunyuan Video and inpainting models. - **Model Formats**: diffusers and ckpt models. - **Training Methods**: Full fine-tuning, LoRA, embeddings. - **Masked Training**: Focuses on specific parts of the samples. - **Automatic Backups**: Regularly back up your training progress. - **Image Augmentation**: Randomly transforms images to create a diverse dataset. - **TensorBoard**: Tracks the training progress with TensorBoard. - **Multiple Prompts per Image**: Train on multiple prompts per image. - **Noise Scheduler Rescaling**: Rescales the noise scheduler from the paper. - **EMA**: Train your own EMA model. - **Aspect Ratio Bucketing**: Automatically trains on multiple aspect ratios. - **Multi-Resolution Training**: Train multiple resolutions simultaneously. - **Dataset Tooling**: Automatically captions your dataset or creates masks for masked training. - **Model Tooling**: Converts between different model formats. - **Sampling UI**: Sample the model during training

### Overview: Token Monitor is a local-first desktop widget that tracks token usage and limits across 46+ AI coding tools, including Claude Code, Codex, Cursor, OpenCode, OpenClaw, and more. It provides real-time multi-device sync, historical usage trends, and detailed session information. ### Key Features: - **Token Usage Tracking**: Shows real-time token usage across various AI tools. - **Limits Check**: Checks and displays account limits for each tool. - **Session Details**: Provides detailed session information including project, model, and device breakdowns. ### Tech Stack: - JavaScript ### Why It Matters: Token Monitor simplifies tracking and managing AI tool usage, making it easier to monitor and manage token usage across multiple tools and devices.

### Overview: Determined is an open-source machine learning platform designed to simplify distributed training, hyperparameter tuning, experiment tracking, and resource management. It works seamlessly with PyTorch and TensorFlow, ensuring faster results, better models, cost-effective resource management, and reproducible experiments. ### Key Features: - **Distributed Training**: Accelerates model training by leveraging multiple GPUs or nodes. - **Hyperparameter Tuning**: Optimizes model performance by adjusting hyperparameters efficiently. - **Resource Management**: Reduces cloud GPU costs by optimizing resource allocation. - **Experiment Tracking**: Enables comprehensive analysis and reproducibility of experiments. ### Tech Stack: Go ### Why it Matters: Determined simplifies the complex process of machine learning research and development, making it accessible and efficient for both small and large teams. Its integration with popular frameworks like PyTorch and TensorFlow ensures compatibility and ease of use, while its robust features and community support make it a valuable tool for deep learning researchers and practitioners. Repository: determined-ai/determined GitHub Stars: 3245 GitHub Description: Determined is an all-in-one deep learning platform compatible with PyTorch and TensorFlow. It simplifies distributed training, hyperparameter tuning, experiment tracking, and resource management. Works with PyTorch and TensorFlow. GitHub Stars: 3245

### Overview: ONNX-TensorRT is a library that parses ONNX models for execution with TensorRT, enabling efficient inference using NVIDIA's TensorRT engine. ### Key Features: - Parses ONNX models for TensorRT backend execution. - Supports full-dimensions and dynamic shape support. - Compatible with TensorRT 11.4 and previous versions. - Includes operator support matrix for ONNX operators. - Provides a list of commonly seen issues and questions. ### Tech Stack: - C++ ### Why It Matters: ONNX-TensorRT simplifies the process of converting ONNX models into TensorRT engines, making it easier to deploy and optimize models for real-time inference.

Overview: Jittor is a high-performance deep learning framework built on Just-in-Time (JIT) compiling and meta-operators. It offers specialized models and powerful op compilers, making it suitable for a wide range of deep learning tasks. Key features: - JIT compiling - Meta-operators - High-performance op compilers - Comprehensive model libraries including image recognition, detection, segmentation, generation, etc. - Python front-end and high-performance language back-end Tech stack: Python Why it matters: Jittor's ability to generate high-performance code with specialized models makes it an ideal choice for deep learning tasks, especially in terms of efficiency and model optimization. Repository: Jittor/jittor GitHub stars: 3239 GitHub description: Jittor is a high-performance deep learning framework based on JIT compiling and meta-operators.

Overview: pstack is a comprehensive skill stack for Claude Code, Codex, Pi, GitHub Copilot, and other agent harnesses, designed to enhance agent outcomes by improving code conciseness, simplicity, and verification. Key features: - pstack integrates with Claude Code, Codex, Pi, GitHub Copilot, and other agent harnesses. - Tracks upstream and carries named policy forks. - Simplifies code by keeping it concise and simple. - Verifies invariants and tracks concurrency bugs with the agent-formal-verify plugin. - Provides rigorous agent workflows with Cursor primitives translated for other harnesses. Tech stack: JavaScript Why it matters: pstack enhances agent outcomes by improving code conciseness, simplicity, and verification, making it a valuable tool for developers looking to streamline their workflows and ensure code quality.

Overview: This repository contains implementations of basic Reinforcement Learning (RL) algorithms with minimal lines of code, focusing on PyTorch. Each algorithm is complete in a single file, with the length of each file limited to 100-150 lines, ensuring that training can be completed within 30 seconds without the need for a GPU. Key features: - REINFORCE (67 lines) - Vanilla Actor-Critic (98 lines) - DQN (112 lines, including replay memory and target network) - PPO (119 lines, including GAE) - DDPG (145 lines, including OU noise and soft target update) - A3C (129 lines) - ACER (149 lines) - A2C (188 lines) - SAC (171 lines) added!! - PPO-Continuous (161 lines) added!! - Vtrace (137 lines) added!! Tech stack: PyTorch Why it matters: This repository provides a concise and efficient way to implement basic RL algorithms, making it ideal for developers looking to quickly get started with RL without the need for extensive code. The minimal lines of code ensure that the algorithms can be trained efficiently, even on low-powered devices.

Overview: Agent Chat UI is a web application that allows developers to interact with LangGraph agents via a chat interface, enabling seamless communication with LangGraph servers. Key features: - Real-time chat functionality - Integration with LangGraph agents via a `messages` key - Support for LangSmith API key authentication - Built-in support for Agent Builder deployments Tech stack: TypeScript Why it matters: This tool simplifies the interaction with LangGraph agents, making it easier for developers to integrate AI capabilities into their applications without the need for complex setup or configuration.

Overview: This repository is a collection of deep learning model converters for various deep learning frameworks and software platforms. It aims to help developers convert models from one framework to another, facilitating model interoperability across different tools and environments. Key features: - Converts models between MXNet, Caffe, Caffe2, ONNX, Theano, PyTorch, TensorFlow, Keras, Darknet, and Chainer. - Provides a comprehensive list of converters from the GitHub model zoo, including official and user-implemented models. - Supports conversion to Core ML for iOS applications. - Offers ONNX support for model export and inference. Tech stack: Python, C++, Java, JavaScript, and various deep learning frameworks. Why it matters: This repository serves as a valuable resource for developers looking to convert models between different deep learning frameworks, ensuring model compatibility and ease of use across various tools and environments.

Overview: Cindy is an open-source AI agent that seamlessly integrates with your projects and apps, working out of the box and ready for customization over time. Key features: - Seamless integration with your projects and apps - Supports multiple harnesses, models, and tools - Runs locally on your machine, using your real files and apps - Native harnesses and models are in development - Models and harnesses can switch mid-task, maintaining continuous workspace, skills, and tools - Can drive browsers, computers, and phones, and take work from IM and schedules - Open-source client for Cindy, including desktop and mobile apps and shared packages Tech stack: TypeScript Why it matters: Cindy is a powerful and flexible AI agent that can be easily integrated into any project, providing real work done on your own computer, without the need for additional setup or configuration. Repository: makecindy/cindy Primary language: TypeScript GitHub stars: 3227 GitHub description: Consider it done. The open-source AI agent that works out of the box · 想到，就能做到。开源、开箱即用的 AI Agent.

### Overview: This repository is a curated collection of essential resources and links related to data science, covering a wide range of topics such as software, platforms, languages, techniques, and more. It serves as a one-stop destination for anyone interested in data science. ### Key Features: - **Software and Platforms**: Links to various software tools and platforms used in data science. - **Languages**: Resources for programming languages commonly used in data science. - **Techniques**: Articles and tutorials on data science techniques and methodologies. - **Data Science Tools**: Links to data science tools and libraries. - **Machine Learning**: Jupyter notebooks for learning and practicing machine learning. - **Artificial Intelligence**: Resources on artificial intelligence, including AI basics and future trends. - **AWS**: Comprehensive guides on Amazon Web Services, including cloud storage, compute, and load balancing. ### Tech Stack: Python, Jupyter Notebook, AWS, Machine Learning, Data Science Tools. ### Why It Matters: This repository provides a comprehensive and up-to-date collection of resources for anyone looking to deepen their knowledge in data science, artificial intelligence, and related fields. It serves as a valuable resource for both beginners and experienced practitioners, offering a wide range of materials to enhance their skills and understanding.

### Overview: OSWorld is a comprehensive benchmarking platform designed for evaluating multimodal agents in open-ended tasks across real computer environments. This project showcases the latest advancements in artificial intelligence and machine learning, providing a robust framework for researchers and developers to test and compare their models. ### Key Features: - **Benchmarking Tasks**: OSWorld offers a diverse set of tasks that simulate real-world scenarios, including image, text, and audio inputs. - **Realistic Environments**: The platform emulates various real-world environments, ensuring that the agents are tested under conditions as close to reality as possible. - **Open-Ended Tasks**: The tasks are designed to be open-ended, allowing agents to learn and adapt to new situations without predefined rules. ### Tech Stack: Python ### Why It Matters: OSWorld provides a crucial tool for researchers to evaluate the performance of multimodal agents in open-ended tasks, thereby advancing the field of AI and machine learning. The platform's ability to simulate realistic environments and tasks makes it an invaluable resource for researchers and developers alike. Repository: xlang-ai/OSWorld GitHub Stars: 3219

### Overview: A local MCP server designed for agent literature work, providing original-LaTeX section reads, BibTeX from arXiv metadata, and topic watches. Papers are stored locally and can be searched optionally. ### Key features: - Original-LaTeX section reads - BibTeX from arXiv metadata - Topic watches - Papers stored on disk - Optional search functionality ### Tech stack: Python ### Why it matters: This server is a local MCP solution that focuses on literature work, providing a seamless experience for researchers and agents. It leverages local storage and avoids external dependencies, making it ideal for environments where external services might be unavailable or unreliable. ### Repository: blazickjp/arxiv-mcp-server ### GitHub stars: 3226 ### GitHub description: A local MCP server for agent literature work. The differentiator is original-LaTeX section reads, BibTeX from arXiv metadata, and topic watches. Papers stay on disk. The working loop is paper ID → outline → one section → citations. Search is optional.

Overview: Playwright Skill is a general-purpose Playwright automation tool designed for coding agents, enabling them to write and execute Playwright scripts effortlessly. Key features: - Any Automation Task: Writes custom code tailored to specific requests, not limited to pre-built scripts. - Visible Browser by Default: Real-time visibility with `headless: false`, enhancing usability. - Portable Executor: Runs file and inline scripts with stable module resolution. - Progressive Disclosure: Concise SKILL.md with full API reference loaded only when needed. - Safe Cleanup: Smart temp file management without race conditions. - Comprehensive Helpers: Optional utility functions for common tasks. Tech stack: JavaScript. Why it matters: This skill is essential for agents needing to write and execute Playwright scripts, providing a convenient and versatile solution for coding agents. Repository: lackeyjb/playwright-skill Primary language: JavaScript GitHub stars: 3221

### Overview: OpenVINO™ Notebooks is a collection of Jupyter notebooks designed to help developers learn and experiment with the OpenVINO™ Toolkit. These notebooks provide an introduction to OpenVINO basics and demonstrate how to leverage the toolkit for optimized deep learning inference. ### Key Features: - Introduction to OpenVINO basics - Hands-on tutorials for using OpenVINO APIs - Practical examples for deep learning inference ### Tech Stack: Jupyter Notebook ### Why It Matters: These notebooks serve as a valuable resource for developers looking to understand and utilize the OpenVINO Toolkit effectively, making it easier to implement optimized deep learning inference in their projects. ### GitHub Stars: 3,226 ### Repository: [openvinotoolkit/openvino_notebooks](https://github.com/openvinotoolkit/openvino_notebooks) ### Primary Language: Jupyter Notebook ### GitHub Description: 📚 Jupyter notebook tutorials for OpenVINO™ ### Table of Contents: - [Table of Contents](#table-of-contents) - [📝 Installation Guide](#-installation-guide) - [🚀 Getting Started](#-getting-started) - [⚙️ System Requirements](#️-system-requirements) - [💻 Run the Notebooks](#-run-the-notebooks) - [To Launch a Single Notebook](#to-launch-a-single-notebook) - [To Launch all Notebooks](#to-launch-all-notebooks) - [🧹 Cleaning Up](#-cleaning-up) - [⚠️ Troubleshooting](#️-

Overview: MASt3R-SLAM is a real-time dense SLAM system that leverages 3D reconstruction priors for improved accuracy and efficiency. Key features: - Real-time dense SLAM - 3D reconstruction priors - Compatible with various datasets and formats Tech stack: Python Why it matters: MASt3R-SLAM offers significant improvements in accuracy and efficiency compared to traditional dense SLAM methods, making it a valuable tool for real-time applications in robotics, computer vision, and autonomous systems.

### Overview: Whisper & Faster-Whisper standalone executables for those who don't want to bother with Python. ### Key features: - **Faster-Whisper**: x86-64 compatible with Windows 7, Linux v5.4, macOS v10.15 and above. Supports Windows 10, Linux v5.15 and above with Faster-Whisper-XXL. - **Faster-Whisper-XXL**: x86-64 compatible with Windows 10, Linux v5.15 and above. Includes experimental settings and tweaks for better performance. - **Whisper**: x86-64 compatible with Windows 10 and above. Requires medium model for decent transcription. - **Standalone Executables**: Downloadable from the Releases section. - **CUDA Detection**: Automatically chooses GPU if CUDA is detected. - **Command-Line Interface**: Compatible with Subtitle Edit, Tero Subtitler, FFAStrans, AviUtl, and PotPlayer. - **Model Defaults**: Models are downloaded automatically or manually from Systran and Purview. - **VAD Methods**: Alternative VAD methods like 'silero_v3', 'silero_v4', 'silero_v5', 'pyannote_v3', 'pyannote_onnx_v3', 'auditok', 'webrtc'. - **Speaker Diarization**: Includes speaker diarization feature. - **Faster-Whisper-XXL Pro**: Special non-public version for donators

### Overview: Ableton MCP is a tool that allows you to control and manipulate your Ableton Live session using Claude AI. It enables you to create, modify, and manipulate MIDI and audio tracks, as well as select instruments and effects from Ableton's library. The system is built on a socket-based server that allows for two-way communication between Claude AI and Ableton Live. ### Key Features: - **Track Manipulation**: Create, modify, and manipulate MIDI and audio tracks. - **Instrument and Effect Selection**: Access and load instruments, effects, and sounds from Ableton's library. - **Clip Creation**: Create and edit MIDI clips with notes. - **Arrangement View Composition**: Build full songs autonomously in Arrangement View, including sections like intro, buildup, drop, and outro. - **Session Control**: Start and stop playback, fire clips, and control transport across Session View and Arrangement View. - **Anonymous Telemetry**: Usage tracking to help improve the tool (can be disabled). ### Tech Stack: Python ### Why it Matters: Ableton MCP provides a powerful way to integrate Claude AI into the music production process, allowing for seamless control and manipulation of your live sessions. This tool can help streamline your workflow, automate repetitive tasks, and create more efficient music production. ### Repository: ahujasid/ableton-mcp - **Primary Language**: Python - **GitHub Stars**: 3211 - **GitHub Description**: Control Ableton Live with any LLM: create tracks, arrange clips & compose music via MCP ###